J . Fluid Mech. (1972). vol. 52, part 4, pp.

609-638 609
Printed in Great Britain

A Reynolds stress model of turbulence and its

application to thin shear flows
By K. HANJALIC? AND B. E. LAUNDER
Mechanical Engineering Department, Imperial College, London

(Received 20 July 1971)

The paper provides a model of turbulence which effects closure through approxi-
mated transport equations for the Reynolds stress tensor w.
and for the turbu-
lence energy-dissipation rate E . I n its most general form the model thus entails
the solution of seven transport equations for turbulence quantities but contains
only six constants to be determined by experiment. It is demonstrated that the
proposed approximation to the pressure-rate-of-strain correlations leads to
satisfactory predictions of the component stress levels in plane homogeneous
turbulence, including the non-equality of the lateral and transverse normal-stress
components.
For boundary-layer flows a simpler version of the model is derived wherein
transport equations are solved only for the shear stress --, the turbulence
energy k, and E . This model has been incorporated in the numerical solution
procedure of Patankar & Spalding (1970) and applied to the prediction of a
number of boundary-layer flows including examples of flow remote from walls,
those developing along one wall and those confined within ducts. Three of the
flows are strongly asymmetric with respect to the surface of zero shear stress and
here the turbulent shear stress does not vanish where the mean rate of strain
goes to zero. I n most cases the predicted profiles and other quantities accord with
the data within the probable accuracy of the measurements.

1. Introduction
Over the past few years a number of general and economical numerical pro-
cedures have been developed for solving the systems of strongly nonlinear partial
differential equations which describe the dynamic behaviour of a viscous fluid.
The arrival of these procedures has had (and continues to have) a two-pronged
influence on research in turbulent flows. First, it has shifted the emphasis of
researoh directly towards prescribing the Reynolds stresses; for the numerical
procedures in question solve the time-averaged equations of motion which contain
the Reynolds stresses as unknowns. Thus, such matters as parametric descrip-
tions of mean-velocity profiles, entrainment formulae or overall mean-flow
energy-dissipationrates (all of which,in the past fifteen years or so, have appeared
as important components of one or more integral procedure for turbulent
boundary layers) have taken on a consequential, rather than causative, role in
Present address : Mdinski Fakultet, Sarajevo, Yugoslavia.
39 F L M 52
610 K . HanjaliC and B . E . Launder
the evolution of the flow. Second, the very fact of being able to solve systems of
coupled partial differential equations has stimulated the development and use of
more elaborate closing approximations for the Reynolds stresses than had
formerly been employed.
Bradshaw, Ferriss & Atwell (1967) provided what was perhaps the first widely
used boundary-layer prediction procedure in which a differential transport
equation was used to determine the shear stress, the dependent variable in
question being the turbulence energy. Nee & Kovaznay (1968) have also pro-
posed a turbulence model which entails the solution of one differential equation;
they employed the Boussinesq ' effective-viscosity' concept with the effective
viscosity itself appearing as the dependent variable of a differential transport
equation.
A limitation of these one-equation models arises from the fact that the turbu-
lence length scale (appropriate to the energy-containing motions) which appears
in the respective turbulence transport equations is prescribed as an algebraic
function of position in the boundary layer. I n practice, the proposed length-scale
functions are not widely applicable; they are not appropriate to the majority of
flows arising outside the laboratory. Indeed, in many circumstances, if one
neglects the effects of convection and diffusion on the length scale one might just
as well stick to Prandtl's (1925) mixing-length hypothesis.
A number of works have sought to provide models of wider applicability by
supplying a transport equation from which the length scale may be determined;
here may be mentioned, for example, the work of Harlow & Nakayama (1968),
Rodi & Spalding (1970), Ng & Spalding (1969), Spalding (1970) and Jones &
Launder (1972). Each of these models provides an equation for the turbulent
kinetic energy in addition to a scale-determining equation. Closure is thus
accomplished through the Prandtl-Kolmogorov formula for the effective turbu-
lent viscosity vT :
VT = k'l,
where k denotes the turbulence kinetic energy and 1 a length scale proportional
to that of the energy-containing motions. Since an equation for the turbulence
energy is solved, it is clearly not essential for the dependent variable of the second
transport equation to be the length scale itself; any variable of the form k'Zb
would be suitable. Thus Ng & Spalding and Rodi & Spalding have used an equa-
tion for the energy-length-scale product while Harlow & Nakayama and Jones &
Launder have preferred the energy dissipation rate, which at high turbulence
Reynolds numbers may be interpreted as k%/Z.
I n a number of instances turbulence models of the two-equation type have
brought accord between experiment and prediction where hitherto there had
been none. For example, the model of Rodi & Spalding correctly predicts the rate
of spread of the plane mixing layer, the plane jet and the radial jet. With the
Prandtl mixing-length model, however, the mixing length must be taken as
7 %, 9 % and 13 % of the width of the respective flows in order that the growth
rate of the shear flows be in agreement with experiment. Likewise, predictions of
Jones & Launder (1972) have indicated that in a severe acceleration the length
scale in the vicinity of a wall is diminished, causing the mean-flow properties of
 Turbulence in thin shear flows 611
the boundary layer to display features more akin to those of a laminar than those
of a turbulent flow. Again, the results are in quantitative agreement with the
available data.
At the time of writing, the predictive capabilities of two-equation representa-
tions of turbulent motion are far from fully exploited. It appears, however, that
there are many kinds of turbulent flow whose satisfactory description will
require a higher order closure of the Reynolds equations than is implied by the
effective-viscosity concept. The circumstances in question are those where
transport effects upon the Reynolds stresses are appreciable. By way of example,
in a wall jet or in flow through an annulus, diffusive transport of shear stress
results in the non-coincidence of the surfaces of maximum velocity and zero
shear stress, a phenomenon which cannot elegantly be incorporated in the notion
of an effective viscosity.
It is the purpose of the present paper to describe the nature and the performance
of a model of turbulence which, because it provides transport equations for the
Reynolds stresses themselves, does not suffer from the limitations of effective-
viscosity models. The proposed model traces its parentage to the early work of
Rotta (1951) (and to the even earlier work of Chou (1945)). I n its most general
form the model provides transport equations for all the Reynolds stresses but,
for the present, experimental comparison has been restricted to two-dimensional
boundary-layer flows wherein the influence of the normal stresses is acknowledged
to be small. The model has therefore been simplified to one where the turbulent
shear stress and the turbulence energy (i.e. half the sum of the normal stresses)
are the only turbulent velocity correlations determined from transport equations.
I n accord with the practice of Jones & Launder (1972),the length scale of turbu-
lence is obtained from the solution of an equation for the kinematic energy-
dissipation rate. This three-equation model of turbulence, which is developed in
$52-4, contains only six empirical constants of which two may be determined
from the observed decay of turbulence in the absence of mean strain.
I n $ 5 a detailed comparison of predictions generated by the model is made with
six substantially different nearly parallel flows, including free shear flows,
external wall boundary layers and flows within ducts. With few exceptions, the
predicted profiles of mean and turbulence quantities agree with the measure-
ments within the probable accuracy of the data.

2. The Reynolds stress equations

For a fluid of uniform density p the exact transport equations for the Reynolds
-
stresses ui ui may be expressed in the form

Dt
Convection Generation Destruction 'Redistribution'
612 K . Hanjalic' and B. E. Launder
Equation (2.1), in common with the remainder of the paper, adopts tensor nota-
tion with repeated suffices indicating summation. Lower and upper case u's
denote fluctuating and time averaged velocity components respectively, p
denotes fluctuating pressures and overbars imply the usual time averaging of the
correlations in question.
I n its present form of course, the Reynolds stress equation is not immediately
employablein a model of turbulent motion; the right-hand side of (2.1) contains a
number of correlations of turbulence quantities for whose determination a closed
path must first be prescribed. Indeed, turbulence models may conveniently be
categorized by reference to their treatment of (2.1). Thus the neglect of the
convective and diffusive transport terms in (2.1) and the algebraic approximation
of the remaining ones leads, under favourable circumstances, to the constitutive
relation between the Reynolds stress and mean rate of strain employed by
effective-viscositymodels. If, alternatively, the transport terms are retained but,
as above, all the unknown correlations are approximated by expressions con-
taining mean velocity gradients, Reynolds stresses and length scales alone, then
the level of closure is of the kind adopted by Rotta (1951), Harlow & Hirt (1969)
and Donaldson (1968). Still higher order closures of the Reynolds stress equation
have been proposed by Chou (1945), Davidov (1961) and Kolavandin & Vatutin
(1969) and entail the provision of a set of transport equations for the triple
velocity correlations "izLj"k and, for the last of the above, for the microscales of
turbulence pertaining to the dissipation terms as well.
The closing approximations which we adopt below place our model within the
same category as Rotta's. We eschewed an effective-viscositymodel because it
was especially our intent to account for transport effects on the stresses. On the
other hand, it seemed inappropriate to adopt as elaborate a treatment of the
triple correlations as Chou's, for example, when these terms are normally of
minor importance compared with the 'redistribution' terms in (2.1) for which
only comparatively primitive simulations have been devised.
The following paragraphs describe the restrictions accepted and the assump-
tions made in order to simplify the stress equations to a practically useful form.
The major limitation is that the model should be applicable only to those flow
regions where the local turbulence Reynolds number is high. Under this con-
dition, it may be presumed that the smallest scales of motion (which are pre-
dominantly responsible for the correlation (aui/axk)/(au,/ax,)) are isotropic.
Consequently one replaces the dissipation term in (2.1) by

The requirement of high Reynolds number also enables the viscous diffusion
term in (2.1) to be dropped. The two remaining diffusion terms, however, cannot
be dealt with so certainly. Let us consider the pressure diffusion term first. A
companion experimental study of flow in an asymmetric plane channel (Hanjali6
& Launder 1972) suggests that c@ZJcEx2 is small compared with the other terms
appearing in the conservation equation for turbulence energy (x2being the co-
 Turbulence in thin shear Jlows 613
ordinate direction normal to the planes). It may not be correct to take this single
result as generally indicative of the unimportance of the pressure diffusion
terms, nevertheless, in the absence of any other firm evidence, this is the assump-
tion which is made; the term is accordingly neglected.
We stated above our opinion that the provision of transport equations for the
triple velocity correlations represented an inappropriate level of closure. It is,
none the less, to the equations for that we turn to arrive at an appropriate
algebraic simulation of the term. I n appendix A it is shown that with certain
assumptions the triple correlations may be replaced by the following form
containing only second-order correlations:

where cs is a constant. The multiplier of (2.3) ( y e ) may be interpreted as a time

scale characteristic of the energy containing and (asis demonstrated in appendix
B) of the diffusing motions.
To complete the simulation of equation (2.1), attention is now given to the
pressure rate-of-strain correlation

This term is commonly referred to as a 'redistribution' term for, in the equations

for the normal stresses (i.e. i = j),it is readily demonstrated that the term acts
so as to diminish the difference between the normal-stress components (Hinze
1959). Following Chou, the exact expression? for the pressure rate-of-strain
correlation at some point xomay be written as
-
13% = 4n
pax, [(3%(2)
'svVOl )' (%)I y
(2)'
ax,ax, +2(2)'
= A,,1 + h,,2' (2.4)
I n the above expression terms without superscripts are evaluated at xo,while
those with a prime superscript are evaluated at (xo+ x).
Equation (2.4)shows that the correlation in question originates from two types
of physical process. The first part is generated from a mutual interaction between
turbulence components, while the second arises from the mean rate of strain and
its interaction with the turbulence.
I n a non-isotropic homogeneousflow with small or zero mean rate of strain only
the first part of the pressure-strain term is significant. Since such a flow will decay
to the statistically more probable isotropic state, the process denoted above by
&, must proceed in such a way as to equalize the normal-stress components and
to diminish to zero the shear stresses. Such reasoning led Rotta (1951) to propose
the following plausible form for the term:
w,:, + $,ill = - c$,(s/k)("Qq- s4, 2 w , (2.5)
t Away from the immediate vicinity of a wall.
614 K . HanjaliC and B. E. Launder
where c + ~is a constant. Later, Rotta (1962) showed that Uberoi's (1957) data
provided support for the above approximation, at least for i = j . Equation
(2.5) is the form adopted here.
Considering next the second part of the pressure-strain term, it is noted that if
the turbulence were homogeneous (with the mean velocity varying linearly at
right angles to its vector direction) the term q5ij, could strictly be re-expressed as

+if, 2 = ( 8q / a z m ) ae, (2.6)

where

and the 6's are the Cartesian components of the position vector x.
I n fact, (2.6) is the form adopted in the present model; it is thus implicitly
assumed that any inhomogeneitiesin the flow do not make a major contribution
to the integral appearing in (2.4). The approximation coincides with Chou's
proposals. It remains, of course, to prescribe a?. Neither Chou nor Rotta pro-
posed a general form for this fourth-order tensor though both drew attention to
the constraints, arising from symmetry and mass conservation, which its
components must satisfy, namely

Moreover, as Rotta (1951) observed, it follows from Green's theorem that

up=2Wi. (2.9)

The form of (2.8)and (2.9)suggested first that a? could be satisfactorily approxi-

mated by a linear combination of Reynolds stresses involving one (or both) of the
velocity components u, or ui. Equations (2.7), (2.8) and (2.9) were then sufficient
to determine the exact form of the tensor, but the result did not agree with the
implications of the data of Champagne, Harris & Corrsin (1970) (towhich we shall
shortly turn). Accordingly, double products of Reynolds stresses were added, the
resultant form satisfying the symmetry requirement of (2.7) being
u p = a=CYlj +/3(UmCYij+ msi, +-arnl + uiulSmj)
-- --
+ (yCYm$CY+?jq6rnl6<j+
lj CYmjSil]) k:+ U ( U . m u j U i U ~ + u m u ~ U i u ~ ) / k
--
+ C+2(UmUiuiUj)/k, (2.10)
where a,/3, y , 7,u and clz are constants. I n principle the application of (2.8) and
(2.9) enables five of the constants to be determined in terms of the sixth. If,
however, u and c + are
~ to be non-zero, (2.9) cannot be satisfied exactly. We there-
fore replaced at one point the correlation w . u i i . by wgk . this
With
substitution, the other constants may be expressed as follows in terms of c $ ~ :
a = ( l O - - S ~ + ~ ) / l l /3
, = -(2-6~$2)/11,
y = - (4- 12~+,)/55, 7 = ( 6 - 1&$2)/55, u = -cc2.
 Turbulence in thin shear Jlows 615
The final form of the simulated equations for Reynolds stress may therefore be
written as
E (--8ij-
Dt

where, for brevity, (& + $ji)2 is used to denote the second part of the pressure-
strain term simulated by (2.6), (2.10) and (2.11).
Equation (2.12) contains three constants (cs,c41and c $ ~ ) which
, must be chosen
by reference to experimental data. It is convenient to defer selection of the
diffusion constant cs until $ 4 ; appropriate values for c$, and c42will however be
indicated now. Uberoi’s (1957) measurements of the decay, in the absence of

- -
C41 G4z (u;- 3k)/2k (q- Qk)/2k (u,”- 3k)/2k -E 3 3
2.8 0.45 0.135 - 0.085 - 0.05 0.27
2.5 0.40 0.14 - 0.08 - 0.06 0.30
Champagne et al. 0.14 - 0.09 - 0.05 0-33
TABLE1. Reynolds stresses in a homogeneous shear flow

mean strain, of an isotropic field towards isotropy indicate (as Rotta (1962) has
noted) that c$, should lie between about 2.6 and 3.0. A value of about 2.5 is
suggested by the more recent data of Tucker & Reynolds (1968). To select ciz
we consider a homogeneous turbulent field in which U, increases linearly with
x2, and U2and U, are zero. For this flow E = - ~ , ( d U l / d x 2 and
) hence (2.12) may
be reduced to the following simple algebraic forms:

(2-#k)/2k = (4 + 1oC$,)/33(C$1- 2C$,), (2.13)

(2- $k)/2k = (1 - 1 4 ~ $ , ) / 3 3 ( ~-$ 2, ~ $ , ) , (2.14)

-
(u:- #k)/2k = - (5 - 4 ~ + ) / 3 3 ( ~ $-, 2~$,), (2.15)

Table 1 compares, for two pairs of values for c41 and c+., the values of the Rey-
nolds stresses given by the above equations with the data of Champagne, Harris
& Corrsin (1970). For either pair of constants, agreement between experiment
and prediction is seen to be satisfactory, including the non-equality of the normal
4 q.
stresses and

3. The rate of dissipation of turbulence energy

For the high Reynolds number flows considered here, the rate of dissipation of
turbulence energy E , is equal to v(aui/axl)2.Following Davidov (1961) and Harlow
616 K . HanjalZ and B. E . Launder
& Nakayama (1968) an exact transport equation for E may be constructed which
for high Reynolds numbers may be expressed as

(iii) (iv) (V)

Our task here (one exactly comparable with that of 8 2) is to provide reasonable
closing approximations of the terms on the right-hand side of (3.1) in terms of
u.iUj,E and the mean rate of strain. Attention is turned first to the generation
term denoted by (i). Since a contraction of the indices yields components of E the
following approximation seems appropriate :

where cE1and Zcl are constants. I n fact, the term containing Zcl vanishes when
(3.2) is multiplied by aq/ax,; thus it need not be considered further.
Rodi (1971) has argued that term (ii),which expresses the generation rate of
vorticity fluctuations through the self-stretching action of turbulence, should
be considered in conjuction with term (iii), representing the decay of the
dissipation rate ultimately through the action of viscosity. Where the Reynolds
number is high enough for an inertial subrange to exist, the sum of terms (ii) and
(iii) may be taken as being controlled by the dynamics of the energy cascade pro-
cess transporting energy from low to high wavenumbers and thus as independent
of viscosity. For dimensional homogeneity it is concluded that

(3.3)

There remain two terms in (3.1) to be considered, both of which express the
influence of diffusional transport processes. Term (iv), which accounts for the
diffusion of E from velocity fluctuations, is treated in a manner analogous to its
counterpart in the stress equations of $2. I n appendix A is it shown that a firm
pruning of the exact equation for 81ukleads to the following result:
- C,k- a€
du, = --u u- (3.4)
8 Lax,.

Term (v) represents the diffusional transport of E by pressure fluctuations. An

exact expression for (ap/ax,)(au,/ax,) can readily be constructed which is similar
in appearance to (2.4).Like (2.4),the resulting expression contains two groups of
terms, one of which arises from an interaction between the mean and turbulent
flow field and the other from a self-interaction of the turbulence. Both terms,
however, contain higher order derivatives of the mean or fluctuating velocity
 Turbulence in thin shear Jlows 617
field than appear in (2.4). It is thus consistent with the level of closure adopted
in the equations for Reynolds stress for these pressure diffusion terms to be
neglected. The final form of the simulated transport equation for the energy-
dissipation rate can thus be expressed as follows:

-E-- - C c 1 - - -Ezc,zc,aui
D
Dt k ax,
€2
CE2-fC,-
k ax, 0
a -k-u
a€ .
8
u-
%, (3.5)

The equation contains three constants which must be assigned appropriate

values; it is this topic that we consider next.
Measurements of the decay of turbulence behind a grid (e.g. Batchelor &
Townsend 1948) have established that at high Reynolds numbers the turbulence
energy level varies inversely with distance from the grid ( k = c x ~ lsay,
, where cis a
dimensional constant particular to the experiment in question). Now for this
particular flow, the transport equation for k may exactly be expressed as
U1dk/dx, = - 8

and hence E = c U , X , - ~It

. is readily verified that for (3.5)to be consistent with this
decay law cC2must equal 2.0.
A further relationship among the constants may be obtained by reference to
the properties of the constant-stress layer adjacent to a wall. I n such a flow,
convective transport is negligible and the production and dissipation rates of
-
turbulence energy are equal, i.e. 8 = - u1u2dU,/dx2 = u9dU11dx2. Equation ( 3 . 5 )
may consequently be written as

;(
- u;-
2)::
(e,l-G,2)+C,-
a
aiC2 [ -d2u11du11
ku2- - 0,
ax; ax,
=

where x2 denotes the direotion normal to the wall and u, the friction velocity.
Moreover, on noting the following experimental properties of the flow:
-
dU,ldx, M ~ , / 0 * 4 2 ~ ,k, M 3 . 5 ~ 9 , U; z 1 * 6 ~ : ,
and recalling that cE2should take the value 2.0, it follows that
Cc1 Z 2 - 3&,. (3.6)

4. A simpler version of the model for boundary-layer flows

In two-dimensionalboundary-layer flows (with z1the primary direction of flow
and x2 the primary direction of velocity gradient) the shear stress -w2
will
usually be the only Reynolds stress to exert significant direct influence on the
mean-flow development. Here we exploit this property of boundary-layer flows
to deduce a form of turbulence model whose visual appearance (and, equally, the
task of incorporating it in a practical calculation procedure) is much simpler than
that of the one presented in $52 and 3 above. There are two basic principles
involved; first, that instead of solving transport equations for each of the normal
stresses, an equation is provided for their sum (strictly, half their sum), the
turbulence energy. Second, the normal-stress terms which remain in the equations
618 K . Hanjalid and B. E. Launder
for kinetic energy, shear stress and energy-dissipation rate are then taken as
proportional to the turbulence energy. The reader may perhaps notice that the
latter principle is very similar to one employed by Bradshaw, Atwell & Ferriss
(1967). These authors, however, relate the shear stress to the turbulence energy
and this practice effectively limits their procedure to external wall boundary
layers wherein the shear stress does not change sign.
Prom (2.12) and the defining equations which precede it it may be deduced that
the form of the simulated shear stress and turbulence energy equations appro-
priate to plane boundary layers is

-
(4.1)

Dk = --zG-aU,
- u ---+cC,-
Dt ,ax,
Besides 8 (for which a transport equation is provided below), (4.1) and (4.2)
2
contain and uTas unknowns. It is appropriate to eliminate these by using the
plane shear layer results presented in table 1; this is the only necessary simplifica-
tion. It will substantially simplify the final form, however, and, arguably, will not
diminish the accuracy of the resultant model if, by the same means, u1u.2 also is
eliminated in the coefficient of aUl/ax, in (4.1) and in the final term in (4.2).
-
Lastly, the term u,u,aG/ax,, which contributes to the shear stress diffusion, is
neglected on the grounds that the diffusion term as a whole is mainly of import-
ante where mean velocity gradients are small, and in these regions u1u2
- auya;,
will ordinarily be much smaller than u T a G / a x , .
Thus, with c+l and c ~ given
, the values 2.8 and 0.45, the following pair of
equations results :

- +c$- a (kz
D u x
-=
Dt
-2.8
(-;- ~ , ~ , + 0 * 0 7 kaul)
ax, ax, 8
-- au,u,)
ax,
, (4.3)

Moreover, for the subclass of flows now under consideration, the transport
equation for the energy dissipation rate, ( 3 4 , becomes
Bs U i i a u , C,,sz
Dt = ccl---- k ax, k (4.5)

t It is of interest to note that if convective and diffusive transport is neglected (4.3)

reduces to: - -
-uluz = o . o ~ ( P / cau,/ax,,
) (4.7)
which is equivalent t o the formula used to relate the shear stress to the mean rate of strain
in many turbulent viscosity models of turbulence (0.g. Spalding 1970; Jones & Launder
1972). Moreover, if aUl/ax2 is eliminated from (4.4)by means of (4.6) the energy equation
(again with transport neglected) maybe manipulated to become - G / k = J(0.07) = 0.27,
which, save for the value of the constant (0.27 instead of 0.3) is identical to Bradshaw,
Ferris & Atwell's relation between stress and energy.
 Turbulence in thin shear flows 619
where the coeecient of the last term arises from replacing by 0.5k (cf. the top
line of table 1).Equations (4.3)-(4.5) together with the mean momentum equa-
tion

comprise the proposed form of the model for boundary-layer flows. The model
contains six constants to be determined from experiment. Two of these (c+, and
c+,) have, for algebraic clarity, already been chosen. For the calculations pre-
sented in $ 5 the remainder have been assigned the values noted in table 2. The
entry ‘ computer optimization’ against the diffusion constants c, and c, means
that for the flows considered in Q 5 many calculations were performed in which
the constants were systematically varied. The values chosen are those which we
believed gave the best overall agreement for the flows considered.

Constant Values Basis for choice

Cd1
2.8 Return to isotropy of distorted
turbulence, Rotta (1962)
C@
0.45 Plane homogeneous shear flow,
Champagne et al. (1970).
cs 0.08 Computer optimization
CEl 1.45 Near-wall turbulence (equation (3.6))
CEZ 2.0 Decay of grid turbulence
C€ 0.13 Computer optimization
TABLE2. The empirical constants

5. Comparison of predictions with experiment

The calculations presented below have all been obtained by solving (4.3)-(4.6)
by means of an adapted version of the finite-difference procedure of Patankar &
Spalding (1970); details are provided by Hanjali6 (1970). The above transport
equations are only appropriate for flow regions where the Reynolds number of
the turbulence is high, and this means, of course, that in performing the calcula-
tions the viscous sublayer region must be excluded. Consequently, for wall
boundary layers boundary conditions are applied near (rather than at) the wall.
Thus, except where otherwise stated, the velocity is matched to Patel’s (1965)
version of the logarithmic law of the wall, the gradient of turbulence energy is
set to zero, the mean momentum equation with convection neglected provides a
formula for the shear stress and the energy-dissipation rate is equated to the
generation rate.
At an axis of symmetry the shear stress is made zero, whereas the gradient
normal to the flow direction of the other dependent variables is set to zero.
Finally, at free boundaries the shear stress is again made zero while the other
variables are determined from the degenerate forms of their respective transport
equations which result when derivatives with respect to x, are set to zero.
The data which were uppermost in our mind when devising the model were our
own measurements of fully developed asymmetric flow in a plane channel
620 K . Hanjalic' and B . E . Launder
1.o

0.8

0.6

- 0.2

0.3 I I I I I I I I I I
0 0.2 0.4 0.6 0.8 1.0
XZlD
FIGURE Velocity and shear stress in asymmetric channel flow (Hanjalib
. " Launder
- - -, experiment, ReM = 56600.
1972). -, predictions, R e M = 62000; 0,

3.0 -
- 0.2

- 0.1
a:
N C
3
.y

- -0.1

- -0.2

I I I I I I I I I . -0.3
0 0-2 0.4 0.6 0.8 1.o
XZlDI

FIGURE 2. Correlation coefficient and turbulence energy in asymmetric channel flow

(Hanjalib & Launder 1972). -, predictions, ReM = 62000; 0,0 , experiment, R e M =
56 000.

(Hanjali6 & Launder 1972). The asymmetry had been introduced by fixing ribs
of square cross-section aligned normal to the flow and pitched 10 rib heights
apart to one wall. Figure 1 shows the calculated and measured distributions of
mean velocity and shear stress across the channel for a Reynolds number Re,
(based on maximum velocity U,, and half the distance between plates) of about
60000. The measured shear stress profile is that deduced from the streamwise
pressure gradient and a Stanton tube measurement of the smooth-wall stress.
 Turbulence in thin shear flows 62 1

0 0.2 0.4 0.6 0.8 1 .o

XZ/D
FIGTJRE3. Turbulence energy balance in asymmetric channel flow. -, predictions.
Experiment: 0, -(GdU,/dxz)D/t&; x , - [ d & ~ ) / d ~ ~ 1; D
0 3/ cD/'$z.
4~

I n the calculations, the mean velocity near the roughened wall has been
matched to the equation

which had been found by Hanjalid & Launder to correlate their data over an
appreciable region near the rough wall. Agreement between calculation and
experiment is satisfactory, including the non-coincidenceof the positions of maxi-
mum velocity and zero shear stress. I n figure 2 the hot-wire data of shear stress
and turbulence energy are compared with calculated profiles. The strongly
asymmetric nature of these profiles is again faithfully reproduced by the pre-
dictions. Points to notice include the extensive region on the rough-wall side of
the duct where (G/IC is )virtually constant, the absence of any such region
near the smooth wall and the non-coincidence of the positions of minimum
kinetic energy and zero stress. All these features are well predicted.
622 K . Hanjalic‘ and B. E . Launder
1.o
I .O
0.9
0.6
0.8
0.2

-s”
0.7 0
-0.2
b- 0 6
- 0.6

0.5
- 1.0
0.4
-1.4
0.3
0 0.2 0.4 0.6 0.8 1 .O
(r - r1)/(rz -TI)

(v--r1)/(r2--rJ
FIGURE 4.Velocity, shear-stress and turbulence energy profiles in symmetric annulus with
small radius ratio. Re = 240000, r11r2 = 0.088. -, predictions; 0, 0,- - -, experiment
(Lawn 1970).

The turbulence energy balance for the above flow is shown in figure 3. The
experimental values for generation and diffusion were measured directly; those
for dissipation were obtained as the closing term. A feature of this asymmetric
channel flow is that the diffusion term is of substantially greater importance than
in a smooth channel. The calculated and measured diffusion fluxes are seen
generally to be in satisfactory agreement.
We have also made calculations of flow in a plane channel with smooth walls.
In this case the predicted profiles of mean velocity and turbulence energy fall
between the experimental data of Comte-Bellot (1965) and Laufer (1951) and
hence may be said to lie within the experimental uncertainty.
 Turbulence in thin shear flows 623

20

10

1c

cll

2c

FIGURE 5. Turbulence energy balance in symmetric annulus. Re = 245000,rl/rg = 0.088.

-, predictions; - - -, - experiment (Lawn 1970); ( I ) , - [=(dUl/dr)l (ra-r1)/u;$,
generation; (Z), - [1/2r(d(u~u2r)/dr)]( ~ ~ - r ~ ) / turbulent
u:~, diffusion; (3),e(rg-rl)/th:l,
dissipation.

Lawn (1970) has made a very careful study of another strongly asymmetric
internal flow: that which arises in a symmetric annulus where the radius of the
core tube is only a small fraction of the outer containing tube (0.088).Figure 4
compares his measured profiles of mean velocity, turbulence energy and shear
stress with predictionst; for all profiles good agreement is displayed.
For the boundary condition on mean velocity near the core tube, the additive constant
in the logarithmic law was taken as 4.3 (rather than Patel's value of 5-45)in agreement
with Lawn's data. A referee has pointed out that, in view of this departure of the velocity
from the universal profile, there is some doubt as to whether the boundary condition used
for kinetic energy (generation = dissipation) is still appropriate. The comparison of
calculations and measurements in figure 5 suggests, however, that this local equilibrium
assumption is still adequate.
624 K . HanjaliC and B. E . Launder

22

20

18

16

14

-
8
a 12

10
I

8 -1 *
N C
-5
6 -2 5
1s
4 -3

2 -4

I I I I I I I I I -5
0 0-1 0.2 0-3 0.4 0-5 0.6 0-7 0.8 0.9 1.0
(r- bl)/@2 - bl)
6. Velocity and shear-stress proflles in annulus with rough core. -, predictions.
FIGURE
Experiment (Lawn & Hamlin): 0 , Re = 71000; 0, Re = 182000.

An interesting difference between the behaviour here and in the rough-smooth

channel emerges on consideration of the relative position of maximum velocity
and zero shear stress. I n figure 1 it was shown that ym lies nearer to the surface
with the higher shear stress than yo;the reverse is found, however, in the smooth
annulus. A turbulence energy balance for the flow is shown in figure 5. Agreement
between the profiles of measured and predicted diffusion rates is not quite as
close as for the plane channel considered earlier. Considering the difficulty of
obtaining measurements of energy diffusion, however, the predictions can be
taken as adequate.
The last of the duct flows to be considered is that of Lawn & Hamlin (1969);
this is the fully developed flow in a symmetric annulus with a roughened core
tube, for a radius ratio of 0.56. The calculated and measured profiles of shear
stress and mean velocity are seen, from figure 6, to be in excellent agreement.
Attention is now turned to developing external wall boundary layers. Kleba-
noff's (1955) measurements of a boundary layer developing in a uniform free
stream are compared with predictions in figures 7 and 8; the momentum thick-
ness Reynolds number is 7700. Agreement with the mean velocity and shear stress
data is good but, for the energy profile, values are 15-20 % higher than those
measured over much of the boundary layer, a discrepancy which is probably
 Turbulence in thin shear JEows 625

0 1 2 3. 4 5 6 7 8
XZ (cm)
FIGURE7. Velocity, shear-stress and turbulence energy profiles in the wall boundary layer
at uniform pressure. -, predictions. 0, 0,- - -, experiment (Klebanoff).

greater than the uncertainty of the experimental data. Figure 7 shows that the
difference can be diminished by increasing the value of c,; the consequent effects
on the velocity and shear stress profiles are not distinguishable on the scale of
the figures. Klebanoff obtained his energy balance by measuring the convective
transport and production of energy directly, by estimating the dissipation rate
from measurements of five of the nine terms in the dissipation term ( a ~ ~ l a and
x~)~,
thus obtaining the diffusive flux as the closing term. It is generally supposed,
however, that his estimate of E was too low by an appreciable amount and that
this led to implausibly largevalues for the diffusionterm; the discrepanoybetween
40 F L M 52
626 K . Hanjalit and B . E . Launder

0 0.2 0.4 0.6 0.8 1.0

.,IS
FIGURE 8. Energy balance in wall boundary layer. -, predictions. Experiment: -- -,
Klebanoff; - . -, Townsend ;

6
(3), E - , dissipation; (4), x 10, convection.
u:,
measurement and predictions shown in figure 8 is certainly consistent with this
supposition. Townsend (1951) has made direct hot-wire measurements of the
energy diffusion in a zero-pressure-gradient boundary layer and it is seen from
figure 8 that these are in much better agreement with the predicted profile.
Figure 9 compares the predicted velocity and shear-stress profiles in a plane
wall jet with the experimental data of Tailland & Mathieu (1967). The shape of
the mean velocity profile, like the displacement of the positions of zero stress and
maximum velocity, is in agreement with measured results. However, the general
level of shear stress is too high, an occurrence which causes the predicted growth
 Turbulence in thin shear flows 627

XZIXZ. +
9. Velocity and shear stress profiles in plane wall jet. -, predictions;
FIGURE
e,
0, experiment (Tailland & Mathieu).

rate of the wall jet to be about 2 0 % greater than measurements suggest. At

present we are uncertain whether these differencesrepresent essential deficiencies
of the present model or merely indicate that the boundary conditions used are
not quite appropriate to the flow in question. For example, as remarked above, we
adopted Patel’s value for the additive constant in the logarithmic law (5-45).
However, in view of the substantial negative gradient of shear stress in the near-
wall region of the flow, it is probable that a larger value would have been appro-
priate. Moreover, there seems at least some doubt as to whether, in the experi-
ments themselves, the free stream was genuinely at rest. Finally, it is remarked
that in the calculations- the -direct influences of the normal stresses have been
neglected; the term - a ( ~ -: ~ ; ) / ahas
x ~been omitted from the mean momentum
equation and correspondingterms have been left out of the turbulence transport
equations. Although these terms certainly do not dominate the flow’s evolution
their omission may have contributed to the differencesbetween the measured and
the predicted growth rates.
Lastly, we show the outcome of applying the model to the prediction of two
free shear flows: the plane jet in stagnant surroundings and the plane mixing
layer. The predicted behaviour of the plane jet is compared in figures 10 and 11
with Bradbury’s (1965) data. Mean velocity and energy profiles are generally in
satisfactory accord but the peak value of the predicted shear stress is some 15 %
less than Bradbury’s measurements. As a result, the generation term in the
energy-balance equation is too small in the neighbourhood of x2+ the distance
from the axis at which the velocity is one half that on the centre-line. Apart from
this defect, the calculated and measured energy balances shown in figure 11are in
satisfactory agreement.
40-2
628 K . Hanjalic' and B. E . Launder

1.0 0.025

0.8 0.02

0.6 0.015
m-
5

0.4 0.01

0.2 0,005

0 0

0.08

0.06

-a
5M0.04
fe

0.02

0
0.4 0.8 1.2 1.6 2.0 2.4
XZ/% +
10. Velocity, shear-stress and energy profiles in a plane jet. -,
FIGURE predictions;
- - -, 0, 0,experiments at two stations (Bradbury).

The calculations for the plane mixing layer shown in figure 12 likewise display
reasonably good agreement with the data of Wygnanski & Fiedler (1970). It is
seen, however, that the predictions of the turbulence energy and shear stress are
displaced relative to the measurements towards the zero-velocity boundary of the
flow, xzo;the displacement is about 15 o/o of the width of the shear flow. A further
difference between measurement and prediction is in the rate of spread of the
mixing layer. Wygnanski & Fiedler's data show the tangent of the angle of
spread to be about 0.20, whereas the calculated value is 0.15. The latter, however,
corresponds with the earlier measurements of Liepmann & Laufer (1967).
 Turbulence in thin shear JEows 629

I
'
I I I
P--0
0.016
4 \
0.012

FI
.-
$ 0.008

0.004

0.004

m
2 O.OOE

0.01;

0.0 1t
0.4 0.8 1.2 1.6 2.0 2.4

X21X2. f
11. Energy balance in a plane jet. -,
FIGURE predictions; - - -, experiment (Brad-
bury) ;

6. Concluding remarks
The preceding paragraphs have drawn comparisons between experimental
data of various quasi-parallel shear flows and computer solutions of the same flow
based upon the model of 3 4.I n many contexts, the range of flows examined would
be considered a wide one and the agreement between prediction and measurement
good.? We should remember, however, that these flows are still appreciably less
I n addition to tho flows considered in 3 4, the reader is reminded that the more general
form of the model presented in §§2and 3 was also consistent with experimental data on the
decay of turbulence behind a grid, the Reynolds stress levels in a plane homogeneous shear
layer and the return (in the absence of mean strain) of distorted turbulence towards
isotropy.
630 K . HanjaliC and B. E. Launder

I I I I I 1 -

0.03 - -

ms
b
0.02 -
*
/
\

0.01 - .

I
0
-0.75 -0.50 -0.25 0 0.25 0.50 0.75
(x2 - %, g)/%,o

12. Velocity, shear-stress and energy profiles in plane mixing layer. -, predictions;
FIGURE
, experimental (Wygnanski & Fielder 1970).

complex than are the majority of industrially important shear flows for which
calculation procedures are needed. For such flows we should need t o employ the
complete stress model rather than the simplified version of 54. In conclusion
therefore, it is perhaps appropriate that the approximations made in $$ 2 and 3 to
procure closure should be rescrutinized.
The simulated terms in the Reynolds stress equations lend themselves to more
direct comparison with experiment than do those in the dissipation equation and,
for this reason, they are considered first. The comparisonsmade with the measure-
ments of kinetic energy diffusion in figures 3 , 5 , 8 and 11 provide substantial
support for the gradient-diffusion hypothesis represented by equation (2.3).
The outcome is not altogether surprising for, as is shown in appendix A, the
 Turbulence in thin shear flows 631
representation follows from the neglect of convective transport of the triple
correlation. Whether such an approximation of diffusive transport will prove
adequate in flows with recirculation remains to be seen. The least controversial
of the approximations concerns the representation of the dissipation terms in the
stress equations, for a t high enough Reynolds numbers the dissipative motions
will assuredly be isotropic. Of course, in flows past solid boundaries there must
inevitably be a region near the surface where the Reynolds number of the turbu-
lence is low and here notions of local isotropy must be discarded. I n these regions,
not withstanding some exploratory calculations by Donaldson (1968), there
remains considerable doubt as to the appropriate modelling of the dissipation
processes (and of the other terms in the stress equations also). We regard this as
an important area for further research.
The experimental evidence in support of Rotta's approximation of the pressure
strain term has been cited in 9 2 and by several other workers. Sufficeit to add here
that it combines the virtues of physical plausibility and algebraic simplicity. The
latter cannot be claimed for our simulation of the second part of the pressure
strain term, but by adopting a relatively general form for the tensor a? we were
able to satisfy a great many kinematic constraints and this has apparently led to a
satisfactory approximation of the term.$ We believe that near a wall, where the
rate of change of mean velocity gradient is large, it will be desirable to improve on
the assumption that X@x, is uniform in the integral of (2.4),at any rate if the
normal stresses are to be well predicted. However when the flow is nearly parallel
t o the wall the normal stresses do not exert much direct influence on the mean
flow development so the matter may be only of academic interest.
Turning now to the equation for dissipation, it may be said that the form chosen
for both the generation and destruction terms is suggested by arguments of local
isotropy and is consistent with the practice adopted in closing the stress equations.
The forms adopted for these terms are, in fact, equivalent to the approximations
of Chou (1945), Davidov (1961) and Daly & Harlow (1970). I n treating the
diffusion of dissipation Davidov provided a closed set of transport equations for
each component "lu,; we decided not t o adopt such an elaborate closure. As is
shown in appendix A, simplification of those equations leads t o the much more
tractable gradient-like representation adopted in 3 3. If t,hisform should prove to
possess too limited validity, the more general version, equation (A7), could be
employed without significantly increasing computation time.
Lastly we draw attention to the absence from our model of any explicit
recognition of the intermittent character of the turbulence near the boundary
of the shear flow with a quiescent stream. Of course, for the engineer, the more

i- O f tho flows where diffusion measurements were available to us, two were fully
developed flows where convection is, by definition, zero and the other two were equilibrium
flows in zero pressure gradients, where convective transport is small.
$ The second part of the pressure-strain correlation is instrumental in causing turbu-
lence-driven secondary flows in straight ducts. Productions of this type of flow thus provide
a very sensitive indicator of the satisfactoriness (or otherwise) of the approximation chosen
for that term. It appears from the calculations of Launder & Ying (1971) that the form
given in 0 2 does predict this occurrence and the offect of turbulence driven secondary flows
in square ducts within the accuracy of present measurements.
632 K . H a n j a l i C a n d B. E. L a u n d e r
of the details of turbulence that can be ignored the better, but it is perhaps
significant that our calculations are in less satisfactory agreement with measure-
ments in those flows where a free-stream boundary is present, particularly where
the stream is at rest. Whether, in practice, the aerodynamicist will ever find it
economically worth while to calculate the time-dependent behaviour of the un-
dulating outer edge of the shear flow seems, however, to be very much a matter of
conjecture.

The research whose outcome is presented in this paper has been sponsored by
the Berkeley Nuclear Laboratories of the CEGB. We are pleased to acknowledge
this support and also the keen interest shown in the work by the Board's staff.
We also wish to record that the work has benefited from conservations with
colleagues at Imperial College, particularly with Mr P. Bradshaw, Dr W. P.
Jones and Mr W. Rodi.

Appendix A. Simulation of the turbulent diffusive transport

Some support for the closure approximations expressed by (2.3) and (3.4) may
be drawn from consideration of the exact transport equations for the correlation
iii question. Let us first consider the equation for triple velocity correlation, which
for high turbulence Reynolds numbers may be written as

I11

We simplify the quadruple correlation I11 by recourse to a proposition of

Millionshtchikov (1941), i.e. that when the triple correlations are small and their
distribution properties do not differ substantially from those of a Gaussian one,
the quadruple correlations may be approximated in terms of the second-order
correlations by the formulae which are strictly fulfilled for the normal law. Thus
it is presumed that
-- -- --
uiui U k u, = uiuj.U k U l f ui U k .u p l+ UiUl. U k U j (A 2 )
and, hence, that the sum of terms I1 and I11 may be written as

From an examination of Hanjalii: & Launder's (1972) measurements of asym-

metric flow in a plane channel we have found that, in their experiments, the right-
hand side of (A 3) was generally of the same sign as and several times larger than
term I in (A I), especially in regions where the triple velocity correlations were of
 Turbulence in thin shear flows 633
importance. For wall boundary layers, at any rate, it is thus reasonable that term
I should be neglected.
There remains the correlation between pressure fluctuations and Reynolds
stress (IV) to be considered. An exact expression for this term may be arrived at
by precisely the same path as that leading to equation (2.4).As Chou (1945) has
shown, the resultant expression contains two types of term, one involving just
fluctuating velocity correlations and the other expressing the influence of the
mean rate of strain on the pressure-stress correlation. It is consistent with our
practice in arriving at (2.6) and (2.10) to approximate the latter process by
groups of terms of the form upu,u,8U,/ax,. We have, however, neglected correla-
tions of this type in term I and we likewise do SO here.? Accordingly the pressure-
stress correlation is assumed to be adequately represented by

the form of the right-hand side being suggested by the corresponding a,pproxi-
mation in the Reynolds stress equation (equation (2.5)).
Finally, on neglect of the convective transport term from (A 1) and with the
substitution of the approximations discussed in the above paragraph, the
following algebraic expression emerges for uiujuk:

which is identical to (2.3) in the main text.

A similar treatment may likewise be applied to the components of the diffusion
of dissipation, uuk(aui/ax1)2,which for brevity we denote by Qk. By applying to
the exact transport equation for Qk approximations at the same level as those
used in $ 3 to close the equation for F , the following form is obtained:

On neglecting the convective transport, ( A 6 ) may be rearranged to give the

following explicit algebraic equation for Qk :

Now, Qk is mainly of importance in $he vicinity of a wall where gradients in

dissipation rates are large. It is thus safe to assume that the last term in square
brackets is small in comparison with the first. Moreover, in boundary-layer flows,
at any rate, the second term in brackets is negligible too, since the subscript k will
denote the direction normal to the main flow, and Uk and its derivatives are
negligibly small. Under these circumstances the equation is adequately approxi-
mated by

which is the form adopted in 3 3.

Though admittedly these pressure-stress terms are not necessarily small even if term
I is.
G34 K . HanjaliC and B. E.Launder

100 0

10-1

10-2

-w

z
w

-6
10-3 n
U

JZID
0 0.047
7 0.108
X 0.146
0 0,240

' +o
X
e 0.50
A 0.75
+ 0.85
O +A
lo-'
+
10-1 loo 10' 10'
K1 4
FIGURE
13. Energy spectra in asymmetric channel.

Appendix B. The dissipation length as normalizing scale for one-dimen-

sional spectra
The analysis of the main text brings into prominence t,he scalar turbulence
quantities lc and E . Indeed, for boundary-layer flows it is these variables (together
with the shear stress - G) which are determined by way of transport equations.
Implicit in the evolution of the forms of these equations is the notion that the
directly influential parts of the turbulent motion can be characterized by a single
length-scale parameter. The scale which suggests itself is the so-called dissipation
length scale 1, defined as lcgle.
To provide some basis for assessing the reasonableness of this assumption,
figures 13,14 and 15 show some of our measurements (Hanjalid & Launder 1972)
of one-dimensional spectra normalized with the dissipation length scale. The flow
geometry, as mentioned in the main text, is a plane channel with one rough and
one smooth surface. This gives rise to two quite dissimilar flows in the vicinity of
 Turbulence in thin shear flows 635

1oo

10-1

10-2

h
w 'v
N

G+ +e
v
N

-6: lo-:
43
-
e
n+o
+
x2/D h
0 0.047
v 0.108
';7 foe
10 I x 0.146
0 0.24;
0 0.50
A 0.75
+ 0.85 +%
lo-' a+rh
+
I
10
I
100
I
101
I
1 oz
x
Klls

FIGURE
14. Shear-stress spectra in asymmetric channel.

the walls dominated by the adjacent surface and, over a central region of the
channel, to a complex flow structure wherein the two wall flows interact. The flow
considered is thus not a simple one and it arguably provides a substantial test of
the single-length-scale hypothesis.
Figure 13 shows the normalized energy spectra & ( K ~ Z ~ ) - ~at seven positions in
the channel from near the smooth surface (x,/D = 0) to near the rough wall
(x2/D= I). Clearly for values of K ~ up Z ~ to about 20 the curves are sensibly
universal. As is seen in figure 14, the shear-stress spectrum displays a roughly
comparable level of universality over the same range of wavenumbers. Here it is
t The normalized spectral density ~ & ( K ~ Z ~ is) defined as

sum
$ij(Kilc) P<j(Kils)/->
where Pii is the un-normalized spectra of u,ujthus
pondingly, for the spectra, of triple correlation
( -
F i j ( ~ l Zds()~ , l , ) =

$ij, z(~llg) ui uj(~1

Is) uL(Kllg)/uiuj ut.
636 K . Hanjalic' and B. E . Launder

-+=Po
10-1 10-2

t -
x 0.24
0 0.50 dc

0
X
0

e
X

0
e

10-4 '

10-1 i=2 100 10' XI02

I I
I
I 1
I
I
lo-' loo i = l 10' 1o2 103

K1 I,
FIGURE
15. Triple correlation spectra in asymmetric channel.

mentioned, however, that the spectrum measured at x2/D = 0.146 (which is

approximately coincident with the surface of zero shear stress) has not been
included. It was left out because $12(~1Z,) took values of opposite sign in different
parts of the spectrum. (The result is attributable to a substantial diffusion of
stress from the rough wall region in a middle range of wavenumbers.) So, while
overall the spectrum is reasonably universal, its profile near the plane of zero
shear stress is quite different from those a t other locations in the flow.
Figure 15 shows normalized spectra of triple correlations representing the
+ $22.1(~lZe)]) and of components of the
diffusion flux of shear stress (i[$12,2(~1Z,)
normal stresses. For these measurements only three positions in the channel were
examined. Here there is more variation from one position t o another than was the
case with the double correlation spectra. Bearing in mind, however, the difficulty
of obtaining accurate spectra of triple correlations, the scatter may be considered
t o be not unreasonably large.

REFERENCES
BATCHELOR, G. K. & TOWNSEND, A. A. 1948 Decay of isotropic turbulence in the initial
period. Proc. Roy. SOC.A 193, 539-558.
BRADBURY, L. J. S. 1965 The structure of a self-preserving turbulent plane jet. J . Fluid
Mech. 23, 31.
BRBDSHAW, P. 1967 The turbulence structure of equilibrium boundary layers. J . Fluid
Mech. 29, 625.
BRADSHAW, P., FERRISS, D. H. & ATWELL,N. P. 1967 Calculation of boundary-layer
development using the turbulent energy equation. J . Fluid Mech. 28, 593.
 Turbulence in thin shear flows 637
CHAMPAGNE,F. H., HARRIS,V. G. & CORRSIN,S. 1970 Exporiments on nearly homo-
geneous turbulent shear flow. J . Fluid Mech. 41, 81-141.
CHOU,P. Y. 1945 On velocity correlations and the solution of the equations of turbulent
fluctuation. Quart. Appl. Math. 3, 31.
COMTE-BELLOT, G. 1965 Ecoulement turbulent entre deux parois parallhles. Publ.
Scienti$ques et Techniques d u Ministere de l’Air, no. 419.
DALY,B. J. & HARLOW, F. H. 1970 Transport equations in turbulence. Phys. FZuids,
13, 2634.
DAVIDOV, B. I. 1961 On the statistical dynamics of an incompressible turbulent fluid.
Dokl. Akad. N ~ u LS.S.S.R. 136,47-50.
DONALDSON, C. DU P. 1968 A computer study of an analytical model of boundary layer
transition. A.I.A.A. Paper, no. 68-38.
HANJALI~, K. 1970 Two-dimensional asymmetric turbulent flow in ducts. Ph.D.
thesis, University of London.
HANJALI~, K. & LAUNDER,B. E. 1972 Fully developed asymmetric flow in a plane
channel. J . Fluid Mech. 51, 301.
HARLOW, F. H. & HIRT, C. W. 1969 Generalized transport theory of anisotropic turbu-
lence. Los Alamos Sci. Lab. University of Calgornia Rep. LA 4086.
HARLOW, F. H. & NAKAYAMA, P. I. 1968 Transport of turbulence energy decay rate
Los Alamos Sci. Lab. University of California Rep. LA 3854.
HINZE,J. 0. 1959 Turbulence. McGraw-Hill.
JONES, W. P. & LAUNDER, B. E. 1972 The prediction of laniinarization with a 2-equation
model of turbulence. Int. J . Heat. Mass Transfer, 15, 301.
KLEBANOFF, P. S. 1955 Characteristics of turbulence in a boundary layer with zero
pressure gradient. N.A.C.A. Rep. no. 1247.
KOLAVANDIN, B. A. & VATUTIN, I. A. 1969 On statistical theory of non-uniform turbu-
lence. Int. Seminar on Heat and Mass Transfer, Herceg Novi, Yugoslavia,
LAUFER,J . 1951 Investigation of turbulent flow in a two-dimensional channel. N.A.C.A.
Rep. no. 1053.
LAUNDER, B. E. & YING,W. M. 1971 Fully-developed turbulent flow in ducts of square-
cross section. Mech. Engng. Dept. Imperial College. Rep. TM/TN/A/ll.
LAWN,C. J. 1970 Application of the turbulence energy equation t o fully developed flow in
simple ducts. C.E.G.B. Rep. RD/B 1575.
LAWN,C. J. & HAMLIN, M. J. 1969 Velocitymeasurements in roughened annuli. C.E.G.B.
Rep. RD/B/N 1278.
LIEPMANN, H. N. & LAUFER,J. 1957 Investigation of free turbulent mixing. N.A.C.A.
Tech. Work, no. 1257.
MILLIONSHTCHIKOV, M. D. 1941 On the theory of homogeneous isotropic turbulence.
C.R. Acad. Sci. S.S.S.R. 32,615-619.
NEE, V. W. & KOVASZNAY, L. S. G. 1968 The calculation of the incompressible turbulent
boundary layers by a simple theory. Conference on Conzputation of Turbulent B o u n d a ~ y
Layers. Stanford University.
~ T GK.
, H. & SPALDING, D. B. 1969 Some applications of a model of turbulence for
boundary layers near walls. Mech. Engng. Dept. Imperial College. Rep. BL/TN/A/14.
PATANKAR, S . V. & SPALDINC, D. B. 1970 Heat and Mass Transfer in Boundary Layers,
2nd edn. Intertext Books.
PATEL, V. C. 1965 Calibration of the Preston tube and limitations on its use in pressure
gradient. J . Fluid Mech. 23, 185-208.
PRANDTL, L. 1925 Bericht uber Untersuchungen zur ausgebildeten Turbulenz. 2.angew.
Math. Mech. 5, 136.
RODI,W. 1971 On the equation governing the rate of turbulent energy dissipation.
Mech. Engng. Dept. Imperial College. Rep. TM/TN/A/14.
RODI,W. & SPALDING, D. B. 1970 A two-parameter model of turbulence and its applica-
tion to free jets. Wurmeund Stofftibertragung, 3, 85-95.
638 K . Hanjalic' and B. E. Launder
ROTTA,J. 1951 Statistische Theorie nichthomogener Turbulenz. 2. Phys. 129, 547-572.
ROTTA,J. 1962 Turbulent boundary layers in incompressible flow. I n Progress in Aero-
nautical Xciences, vol. 2. (ed. A. Ferri, D. Kuchemann & L. H. G. Sterne), pp. 1-221.
Macmillan.
SPALDING, D. B. 1970 The prediction o f two-dimensional steady turbulent flows. Mech.
Engng Dept., Rep. EF/TN/A/16.
TAILLAND, A. & MATHIEU, J. 1967 Jet Parietal. J . Mdcan. 6, 103-131.
TOWNSEND, A. A. 1951 The structure of the turbulent boundary layer. Proc. Camb. Phil.
SOC. 47, 375.

TUCKER,H. J. & REYNOLDS, A. J. 1968 The distortion o f turbulence by irrotational

plane strain. J . Fluid Mech. 32, 657.
UBEROI,M. S. 1957 Equipartition o f energy and local isotropy in turbulent flows.
J . App1. Phys. 28, 1165-1170.
WYGNANSKI, I. & FIEDLER,H. E. 1970 Two-dimensional mixing region. J . Fluid Mech.
41. 327-363.